mul 16x16

yo4hsp · Joined: 12 Sep 2008 Posts: 7

Hello,

For my FIR filter, I need a multiplier. I use pic18f2455 and I found on his datasheet (pp 98). This is asm code for hardware multiplier.

yo4hsp · Joined: 12 Sep 2008 Posts: 7

I need a 16x 16 multiplier with sign. For up side code, the formula is:

FvM · Joined: 27 Aug 2008 Posts: 2337 Location: Germany

Generally, you can expect your C compiler to know how to imlement signed int16 mupltiply.
What's the purpose of a DIY assembler implementation?

Guest · Posted: Mon Nov 09, 2009 2:14 am

Hello,

I need to multiply two large numbers with sign (on 16 bits), and the
results will be on 32 bits. I try the hardware multiply (pic18f2455
datasheet) and I receive lots of errors from unrecognized mnemonics.
I used pic18f2455.

bungee- · Joined: 27 Jun 2007 Posts: 206

Like:

Ttelmah · Guest

var = _mul(x,y);

This is a built in function, that multiplies two 8 bit values, producing a 16bit result, or two 16bit values, producing a 32 bit result - exactly what you want... Smile

A lot more efficient than using 32bit internal arithmetic.

Best Wishes

yo4hsp · Joined: 12 Sep 2008 Posts: 7

ok, but "var = _mul(x,y);" is for unsigned numbers. in my filter i have a lots of coefficients and samples signed. whas i have to use??? so, i need "mul" or anything like multiplier for signed numbers with 16 bits, and results on 32 bits...

thanks,

Ttelmah · Guest

If you look at your assembler example, all it does is test the top bits to verify the possible results that give a signed result.

Best Wishes

Guest · Posted: Mon Nov 09, 2009 2:34 pm

Thanks, I see the testing for sign:
...
(-1 • ARG2H<7> • ARG1H:ARG1L • 2^16) +
(-1 • ARG1H<7> • ARG2H:ARG2L • 2^16)

I hope is another function for multiply with signed numbers like "_mul" for unsigned.

ckielstra · Joined: 18 Mar 2004 Posts: 3680 Location: The Netherlands

I think you missed the point. Ttelmah gave you a hint to solve your problem.

Let me establish a bit more: It is basic mathematics to create signed multiply when you have a routine for unsigned multiply. Only the couple of if-statements from your assembly program...

Ttelmah · Guest

and, as a further hint, the CCS #bit directive, gives an 'assembly efficient' way of testing/clearing individual bits in a variable...

Best Wishes

yo4hsp · Joined: 12 Sep 2008 Posts: 7

thanks..

here my code, butt is still with errors. i think my mcu is slow for my application.

signed int32 inmultit;

signed int32 inmultire (signed int16 ARG1H, int16 ARG1L, signed int16 ARG2H, int16 ARG2L)
{ int32 PRODH = 0;
int32 PRODL = 0;
int32 PRODN = 0;
int32 PRODM = 0;

PRODH = _mul( (int16)ARG1H, (int16)ARG2H); //primii 2 octeti (0 si 1) va fi pe 16 biti
PRODM = _mul( (int16)ARG1H, (int16)ARG2L); //cei 2 octeti de mijloc (1 si 2)
PRODN = _mul( (int16)ARG1L, (int16)ARG2H); //cei 2 octeti de mijloc (1 si 2)
PRODL = _mul( (int16)ARG1L, (int16)ARG2L); //ultimii 2 octeti (2 si 3) va fi pe 16 biti

inmultit = _mul( (int16)PRODH, (int16)0x10000) + _mul( (int16)PRODM, (int16)0x100) + _mul( (int16)PRODN, (int16)0x100 ) + (int16)PRODL
- _mul( ( _mul (!bit_test(ARG1H, 3), (int16)ARG2H) ), (int16)0x10000)
- _mul( ( _mul (!bit_test(ARG1H, 3), (int16)ARG2L) ), (int16)0x100 )
- _mul( ( _mul (!bit_test(ARG2H, 3), (int16)ARG1H) ), (int16)0x10000)
- _mul( ( _mul (!bit_test(ARG2H, 3), (int16)ARG1L) ), (int16)0x100 );

return inmultit;
}

///////------------------------------------------------------------------------------

ckielstra · Joined: 18 Mar 2004 Posts: 3680 Location: The Netherlands